China's grand AI-education experiment — and why your kid is already in one

A piece from MIT Technology Review has been making the rounds again — its deep dive into how China set out to run a national experiment in AI-driven education. The detail that tends to stick with people is the scale. Not a pilot in one district. A country deciding, more or less as policy, that algorithms should help shape what millions of children learn next.

It is easy to read that as a story about somewhere else. A different system, a different government, a different set of trade-offs you are glad you do not have to make. That reading is comfortable, and it is wrong.

The experiment is not only “over there”

Strip away the national-policy framing and look at what the technology actually does. A company like Squirrel AI built tutoring software that breaks subjects into thousands of small pieces, watches how a child answers, and adjusts what comes next. Get something right, move on. Stumble, and the system loops back. By the reporting, it grew to roughly two thousand learning centers and signed up something on the order of a million students in just a few years.

Now look at your own living room. The app that decides which video plays next. The game that tunes its difficulty so your child neither quits nor wins too easily. The homework helper that produces a confident paragraph on demand. These are the same underlying machinery — software that models a child and adjusts itself to keep them engaged.

The difference is not whether your child is inside an AI-shaped learning experiment. They already are. The only open question is which kind.

Two versions of the same technology

There is a version of adaptive software that is genuinely good for a child. It meets them where they are, offers a harder problem when they are ready, slows down when they are lost, and quietly removes the boredom that comes from a classroom moving at the speed of the middle. Used well, that is closer to what a patient tutor does than anything a class of thirty can offer.

And there is a version optimized for a different goal entirely: time on screen. That software is also personalized, also responsive, also “adaptive.” It just adapts toward keeping your child watching rather than toward keeping them learning. From a few feet away the two can look identical. A child leaning into a screen, absorbed, responding to something tuned precisely to them.

The thing that separates them is not the algorithm. It is what the algorithm was pointed at. One is built to make a child more capable. The other is built to make a number go up.

The pedagogy is the whole game

This is the part worth sitting with, because it cuts against the usual debate. Most arguments about AI and kids are arguments about the technology — is it good, is it bad, should we allow it, should we ban it. China’s experiment, whatever you make of it, settles the only question that argument was ever really about. The technology is arriving. It is already here, in your house, on the devices your child uses every day.

So the useful question is not whether a child should learn alongside capable machines. It is how — and around what purpose. A tool aimed at engagement will train a child to be engaged. A tool aimed at thinking, building, and getting things wrong on the way to getting them right will train something far more durable.

Those durable things are the skills we keep coming back to: thinking clearly enough to notice when a confident answer is wrong, staying with a problem that has no page number, explaining an idea so another person sees it, making something that was not there before, and handling powerful tools as the one in charge rather than the one being optimized. None of these come from the software by default. They come from what the software is asked to do.

That is the bet underneath Globeskool. We use adaptive technology because, used properly, it can do something a crowded classroom cannot — teach one child at a time, at their own pace, through real projects rather than worksheets. But the technology is the easy part. The deliberate part is pointing it at a child becoming capable instead of a child staying hooked. China’s experiment is a useful reminder that this choice is not theoretical, and not far away. It is being made right now, by whatever your child happens to open next.

See what your child is actually learning to do →